Pancreatic Cancer Analysis Report

1 Introduction

Pancreatic cancer is one of the most lethal and malignant cancer worldwide. It has a five-year survival rate of less than 10% (Chen, 2023) because of the limited treatment options and late diagnosis. For all the pancreatic cancers, pancreatic adenocarcinoma (PAAD) accounts for the majority of all the pancreatic cancer cases. It is challenging to detect it early, and thus impede the therapeutic intervention later on. We believe that understanding the molecular mechanisms behind PAAD is important for developing diagnostic markers and therapeutic targets to improve patients’ care(Wang, 2021).

We aims to identify genes that are differentially expressed between normal pancreatic tissue and PAAD tissue. We will be using genetic data from The Cancer Genome Atlas (TCGA) and the Genotype-Tissue Expression (GTEx) project. By using the bioinformatics approaches mentioend above, we seek to unravel main biological processes, pathways, and molecular functions associated with PAAD development and progression.

Our research questions are as below:

Primary research questions:

1. What genes are differentially expressed between normal pancreatic tissue(GTEx) and pancreatic adenocarcinoma tissue(TCGA)?

2. Which biological processes and pathways are enriched in the differentially expressed genes?

3. How do these findings contribute to our understanding of the molecular mechanisms underlying pancreatic adenocarcinoma?

Secondary research question:

1. Are there other interesting factors like Diabetes, Alcohol Use, Sex that affect gene expression differentially?

With data preprocessing, differential gene expression analysis(DEGs), pathway enrichment analysis(GSEA), and network inference(PANDA), this project will provide insights for us into the biology of PAAD. By stratifying data based on sex, diabetes status,and alcohol consumption, we further plan to investigate molecular features that may refine our understanding of PAAD heterogeneity and its clinical implications.

2 Primary Reseasrch Methods & Result: TCGA VS GTEx

2.0.1 Data Sources

• TCGA (PAAD): Pancreatic adenocarcinoma samples.
• GTEx: Normal tissue samples.

2.0.2 Tools and Packages

• Differential expression analysis was conducted using limma and voom.
• Gene Ontology (GO) and KEGG pathway enrichment analyses were performed.

library(Biobase)
library(limma)
library(ggfortify)
library(biomaRt)
library(gplots)
library(SummarizedExperiment)
library('ggplot2')      # For plotting
library('reshape2')     # For data processing
library('visNetwork')
library('fgsea')

2.0.3 Exploratory Data Analysis (EDA)

• Sample Details:
  • TCGA (179 cancer samples) with gene information for 34,032 genes.
  • GTEx (349 normal samples) with gene information for 31,530 genes.
  • Includes metadata on sex, age, diabetes, and alcohol history.

## Load Data and Preprocess

# downloading data
gtex <- readRDS("gtex_pancreas.rds")
tcga <- readRDS("tcga_paad.rds")

# structure of both datasets
gtex
tcga

• Processing Steps:
  1. Batch effect detection.

# checking sequencing platform
unique(colData(tcga)$tcga.gdc_platform)
names(colData(tcga))[1:5]

2. Calculated TPM (transcripts per million) of genes.

raw_counts <- assay(tcga, "raw_counts")
gene_lengths <- rowData(tcga)$bp_length/1000
rpk <- raw_counts/gene_lengths
scaling_factors <- colSums(rpk)
tpm <- sweep(rpk, 2, scaling_factors, FUN = "/") * 10^6
tpm[1, 0:5]

3. Focused analysis on early stages (1–2).

# filtering TCGA data to only include first and second grade tumors
keep_stages <- tcga@colData@listData$tcga.gdc_cases.diagnoses.tumor_stage %in%
  c("stage i", "stage ia", "stage ib", "stage iia", "stage iib")
dim(tcga)
tcga <- tcga[, keep_stages]
dim(tcga)

# defining a column that outlines the source: GTEX or TCGA
colData(gtex)$source <- "GTEX"
colData(tcga)$source <- "TCGA"

4. Merge normal/Tumour genes, filtered genes with low counts.

## Analysis for Normal vs Tumour

# intersection of columns (features of sample data)
common_cols <- intersect(names(colData(gtex)), names(colData(tcga)))

# identification of common genes for both GTEX and TCGA
common_genes <- intersect(rownames(assay(gtex, "logtpm")),
                          rownames(assay(tcga, "logtpm")))

# updating datasets by applying common genes and common columns filters
gtex_upd <- gtex[common_genes, ]
tcga_upd <- tcga[common_genes, ]
colData(gtex_upd) <- colData(gtex_upd)[, common_cols, drop = FALSE]
colData(tcga_upd) <- colData(tcga_upd)[, common_cols, drop = FALSE]

# merging
merged_logtpm <- cbind(assay(gtex_upd, "logtpm"), assay(tcga_upd, "logtpm"))
dim(merged_logtpm)
merged_colData <- as.data.frame(rbind(colData(gtex_upd), colData(tcga_upd)))
dim(merged_colData)

# genes will be filtered if they have very low counts (tpm < 1) in 25% of the samples
merged_exprs <- cbind(assay(gtex_upd, "raw_counts"), assay(tcga_upd, "raw_counts"))
merged_gene_lengths <- rowData(gtex_upd)$bp_length/1000
merged_rpk <- merged_exprs/merged_gene_lengths
merged_scaling_factors <- colSums(merged_rpk)
merged_tpm <- sweep(merged_rpk, 2, merged_scaling_factors, FUN = "/") * 10^6
keep <- rowSums(merged_tpm > 1) >= ncol(merged_tpm)/4
gtex_upd <- gtex_upd[keep, ]
tcga_upd <- tcga_upd[keep, ]
saveRDS(gtex_upd, file("gtex.rds"))
merged_exprs <- merged_exprs[keep, ]
sum(keep)

5. Conducted principal component analysis (PCA).

PCA <- prcomp(t(assay(tcga_filtered, "logtpm")))
p <- autoplot(PCA, data = as.data.frame(colData(tcga)),
              colour="tcga.gdc_cases.demographic.gender")
p
PCA <- prcomp(t(merged_logtpm[keep,]))
p <- autoplot(PCA, data = as.data.frame(rbind(colData(gtex_upd[keep,]), colData(tcga_upd[keep,]))), colour="source")
p

PCA <- prcomp(t(merged_logtpm))
p <- autoplot(PCA, data = merged_colData, colour="source")
p

knitr::include_graphics("./PCA.png")

2.0.3.1 PCA Interpretation

We can see the GTEx and TCGA are clearly separated into two clusters, especially in the PC1 Axis. This indicates that there is a difference in the factors explaining the variation in these datasets. Since both datasets contained samples generated from the Illumina platform only, further differential expression analysis will likely reveal true biological differences between these samples instead of batch effects that may be attributed to the choice of platform itself.

2.1 Differential Expression Analysis

2.1.1 Voom Mean-variance trend

knitr::include_graphics("./Voom.png")

2.1.1.1 Voom Interpretation:

The graph is as expected by statistical theory of variance, where variance decreases as count size increases.(UC Davis Bioinformatics, 2018)

2.1.2 Mean Variance trend

fit <- lmFit(voomOutput, design)
fit <- eBayes(fit)
plotSA(fit, main="Final model mean-variance trend")

knitr::include_graphics("./main_mean_var.png")

2.1.2.1 Mean Variance trend Interpretation

It indicates that there are no concerning patterns in our data as the data was appropriately transformed for linear analysis.

2.1.3 Volcano Plot

knitr::include_graphics("./Volcano.png")

2.1.3.1 Volcano Interpretation:

The graph does not show any concerning pattern and follow the lecture example. Plot shows large number of differentially expressed genes between the TGCA and GTEX, and specifically, more differentially expressed genes appear to be upregulated than downregulated.

2.1.4 MA Plot

knitr::include_graphics("./MA.png")

2.1.4.1 MA plot Interpretation:

The graph does not show any concerning pattern and follow the lecture example. Those genes that displayed the greatest positive log fold change (upregulated in cancer samples) had relatively low expression in both samples, whereas those genes that displayed the most negative log fold change (downregulated in cancer samples) were observed as having a broad range of average expressions levels.

2.2 GO and KEGG Enrichment Analysis

2.2.1 Gene Ontology (GO)

2.2.1.1 Over Expressed GO terms in TCGA

# Perform over-representation analyses for Gene Ontology terms using the limma package
go_tcga <- goana(up_tcga_ID, species="Hs")
# This is the list of top GO terms enriched for genes highly expressed in tcga
topGO(go_tcga, ontology = "BP")

knitr::include_graphics("./over-tcga.png")

The GO terms can be grouped into key functional categories related to pancreatic cancer biology.

Immune System and Defense – Overexpressed terms include immune system process (GO:0002376), regulation of immune system process (GO:0002682), immune response (GO:0006955), defense response (GO:0006952), leukocyte activation (GO:0045321), positive regulation of immune system process (GO:0002684), and cell activation (GO:0001775). Tumors manipulate immune responses to promote growth and evade detection(Wikipedia, n.d.), contributing to an inflammatory microenvironment, a hallmark of pancreatic cancer.

Cell Adhesion and Migration – Key terms include cell adhesion (GO:0007155), regulation of cell adhesion (GO:0030155), cell-cell adhesion (GO:0098609), positive regulation of cell adhesion (GO:0045785), and cell migration (GO:0016477). Tumor progression and metastasis involve epithelial-to-mesenchymal transition (EMT), disrupting adhesion and enhancing cell motility and invasiveness.

Stimulus Response and Regulation – This category includes response to external stimulus (GO:0009605), response to stimulus (GO:0050896), regulation of response to stimulus (GO:0048583), positive regulation of response to stimulus (GO:0048584), positive regulation of biological process (GO:0048518), and positive regulation of cellular process (GO:0048522). Overexpression of these genes in pancreatic adenocarcinoma (PAAD) aids tumor survival by promoting immune evasion, adaptation to hypoxia, and resistance to therapy, while driving metastasis through EMT.

Multicellular Organism Processes – Terms include multicellular organismal process (GO:0032501) and regulation of multicellular organismal process (GO:0051239). These pathways regulate tumor growth, tissue remodeling, and interactions between cancer cells and the microenvironment. Overexpression supports proliferation, immune modulation, and communication with stromal cells, facilitating tumor survival and metastasis.

This clustering highlights processes driving tumor progression, immune evasion, and metastatic behavior in pancreatic cancer.

2.2.1.2 Over Expressed GO terms in GTEx

go_gtex <- goana(up_gtex_ID, species="Hs")
# This is the list of top GO terms enriched for genes highly expressed in gtex
topGO(go_gtex, ontology = "BP")

knitr::include_graphics("./over-gtex.png")

The GO terms can be grouped into key functional categories related to normal pancreatic tissue biology.

Metabolism and Catabolism – Overexpressed terms include alpha-amino acid catabolic process (GO:1901606), carboxylic acid catabolic process (GO:0046395), organic acid catabolic process (GO:0016054), proteinogenic amino acid catabolic process (GO:0170040), amino acid catabolic process (GO:0009063), L-amino acid catabolic process (GO:0170035), small molecule catabolic process (GO:0044282), alpha-amino acid metabolic process (GO:1901605), amino acid metabolic process (GO:0006520), oxoacid metabolic process (GO:0043436), and organic acid metabolic process (GO:0006082). These pathways reflect the high metabolic activity required to maintain homeostasis and nutrient processing in normal pancreatic tissues, supporting energy production and normal cellular function. In contrast, cancer cells reduce these catabolic and metabolic processes as they shift towards anabolic pathways to fuel rapid proliferation and tumor growth.

Digestion and Circulation – Key terms include digestion (GO:0007586), circulatory system process (GO:0003013), and blood circulation (GO:0008015). These processes are critical for maintaining normal pancreatic functions, such as enzyme secretion(like insulin) and nutrient absorption. In normal tissues, digestion and circulation ensure homeostasis and energy supply. However, cancer cells downregulate these functions as they prioritize growth and survival over normal physiological activities.(Wikipedia, n.d.)

Homeostasis and Transport – This category includes chemical homeostasis (GO:0048878), metal ion transport (GO:0030001), and import across plasma membrane (GO:0098739). Normal cells regulate ion transport and maintain chemical balance to support stable internal environments and proper enzyme activity in the pancreas. Cancer cells disrupt these pathways to promote proliferation and invasion, resulting in lower expression of these homeostatic functions.

Amino Acid and Nitric Oxide Regulation – Terms include cellular modified amino acid metabolic process (GO:0006575), cellular modified amino acid catabolic process (GO:0042219), and regulation of nitric oxide mediated signal transduction (GO:0010749). In normal cells, these processes contribute to metabolic balance and cellular communication. Pancreatic cancer cells downregulate these pathways as disrupted signaling and metabolic reprogramming favor tumor progression, immune evasion, and survival.

This clustering highlights the metabolic, circulatory, and homeostatic processes essential for normal pancreatic function, contrasting with the metabolic rewiring and disrupted physiological pathways characteristic of pancreatic cancer cells.

2.2.2 KEGG Pathways

2.2.2.1 Over expressed TCGA

# Perform over-representation analyses for KEGG pathways
kegg_tcga <- kegga(up_tcga_ID, species="Hs")
# This is the list of top KEGG pathwayss enriched for genes highly expressed in tcga
topKEGG(kegg_tcga)

knitr::include_graphics("./kegg-tcga.png")

Interpretation:

The top overexpressed KEGG pathways in pancreatic cancer shows key processes driving tumor growth, immune evasion, and metastasis. Phagosome, cytokine-cytokine receptor interaction, and cell adhesion molecules pathways shows an inflammatory microenvironment and enhanced immune signaling that promote tumor survival and immune suppression. ECM-receptor interaction and focal adhesion(Dafrazi, 2023) pathways indicate increased cell migration and invasion, for metastasis. These pathways indicate how pancreatic tumors manipulate immune responses and cellular interactions to sustain growth and spread.

2.2.2.2 Over Expressed GTEx

# Perform over-representation analyses for KEGG pathways
kegg_gtex <- kegga(up_gtex_ID, species="Hs")
# This is the list of top KEGG pathwayss enriched for genes highly expressed in gtex
topKEGG(kegg_gtex)

knitr::include_graphics("./kegg-gtex.png")

Interpretation:

The top overexpressed KEGG pathways in normal pancreatic tissue reflect essential functions for digestion, secretion, and metabolic regulation. Pancreatic secretion, protein digestion and absorption, and fat digestion pathways shows the pancreas’s role in enzyme production and nutrient processing, which are reduced in tumor cells that prioritize growth over normal function. Glycine, serine, threonine metabolism, and one-carbon pool by folate pathways indicate active amino acid and folate metabolism, supporting cellular maintenance and DNA synthesis. Carbohydrate and starch metabolism reflects energy utilization, which is often disrupted in tumor cells that shift toward anabolic pathways. These pathways demonstrate the pancreas’s role in maintaining metabolic homeostasis, which is lost as cancer progresses. It corroborates with our findings in the GO analysis.

2.3 Network Analysis

2.3.1 PANDA Networks

male_network_full <- read.table("output_males.txt", header=TRUE, stringsAsFactors=FALSE)
female_network_full <- read.table("output_females.txt", header=TRUE, stringsAsFactors=FALSE)

2.3.1.1 PANDA-TCGA NETWORK

## This is the TCGA network
edges <- tcga_network
colnames(edges) <- c("from", "to", "value")
nodes       = data.frame(id = unique(as.vector(as.matrix(edges[,c(1,2)]))), 
                    label=unique(as.vector(as.matrix(edges[,c(1,2)]))))
nodes$group = ifelse(nodes$id %in% edges$from, "TF", "gene")
net <- visNetwork(nodes, edges, width = "100%")
net <- visGroups(net, groupname = "TF", shape = "triangle",
                 color = list(background = "purple", border="black"))
net <- visGroups(net, groupname = "gene", shape = "dot",       
                 color = list(background = "teal", border="black"))
visLegend(net, main="Legend", position="right", ncol=1) 

knitr::include_graphics("./panda-tcga-network.png")

2.3.1.2 PANDA CANCER VS NORMAL

## This plots the edges that change the most in cancer vs normal
edges <- diffnet
colnames(edges) <- c("from", "to", "value")
edges$arrows = "to"   
edges$color  = ifelse(edges$value > 0, "green", "red")
edges$value  = abs(edges$value)
edges <- edges[order(edges$value, decreasing = TRUE), ]
edges <- edges[1:200,]
nodes       = data.frame(id = unique(as.vector(as.matrix(edges[,c(1,2)]))), 
                    label=unique(as.vector(as.matrix(edges[,c(1,2)]))))
nodes$group = ifelse(nodes$id %in% edges$from, "TF", "gene")
net <- visNetwork(nodes, edges, width = "100%")
net <- visGroups(net, groupname = "TF", shape = "triangle",
                 color = list(background = "purple", border="black"))
net <- visGroups(net, groupname = "gene", shape = "dot",       
                 color = list(background = "teal", border="black"))
visLegend(net, main="Legend", position="right", ncol=1) 

knitr::include_graphics("./panda-cancer-vs-normal.png")

Plot interpretation:

In the plot above, green arrows mean UP regulated in cancer, red arrows mean DOWN regulated in cancer and we can see a few genes are significantly UP regulated in cancer indicates that pancreatic cancer might replies significantly on certain genes’ function. As we mentioned above, certain functions in GO analysis are specifically over-expressed in cancer such as Immune system and defense, cell adhesion etc, we suspect the genes with thick green arrows are related to some functions that cancer heavily relies on.

2.3.1.3 Differentially Expressed Pathways

## This plots a bubble plot for the differentially expressed pathways


dat <- data.frame(fgseaRes)
# Settings
fdrcut <- 0.05 # FDR cut-off to use as output for significant signatures
dencol_neg <- "blue" # bubble plot color for negative ES
dencol_pos <- "red" # bubble plot color for positive ES
signnamelength <- 4 # set to remove prefix from signature names (2 for "GO", 4 for "KEGG", 8 for "REACTOME")
asp <- 3 # aspect ratio of bubble plot
charcut <- 100 # cut signature name in heatmap to this nr of characters
a <- as.character(dat$pathway) # 'a' is a great variable name to substitute row names with something more readable
for (j in 1:length(a)){
  a[j] <- substr(a[j], signnamelength+2, nchar(a[j]))
}
a <- tolower(a) # convert to lower case (you may want to comment this out, it really depends on what signatures you are looking at, c6 signatures contain gene names, and converting those to lower case may be confusing)
for (j in 1:length(a)){
  if(nchar(a[j])>charcut) { a[j] <- paste(substr(a[j], 1, charcut), "...", sep=" ")}
} # cut signature names that have more characters than charcut, and add "..."
a <- gsub("_", " ", a)
dat$NAME <- a
dat2 <- dat[dat[,"padj"]<fdrcut,]
dat2 <- dat2[order(dat2[,"padj"]),] 
dat2$signature <- factor(dat2$NAME, rev(as.character(dat2$NAME)))
sign_neg <- which(dat2[,"NES"]<0) #Up regulated in cancer
sign_pos <- which(dat2[,"NES"]>0)
signcol <- rep(NA, length(dat2$signature))
signcol[sign_neg] <- dencol_neg # text color of negative signatures
signcol[sign_pos] <- dencol_pos # text color of positive signatures
signcol <- rev(signcol) # need to revert vector of colors, because ggplot starts plotting these from below
g<-ggplot(dat2, aes(x=padj,y=signature,size=size))
g+geom_point(aes(fill=NES), shape=21, colour="white")+
  theme_bw()+ # white background, needs to be placed before the "signcol" line
  xlim(0,fdrcut)+
  scale_size_area(max_size=10,guide="none")+
  scale_fill_gradient2(low=dencol_neg, high=dencol_pos)+
  theme(axis.text.y = element_text(colour=signcol))+
  theme(aspect.ratio=asp, axis.title.y=element_blank()) # test aspect.ratio

knitr::include_graphics("./panda-bubble.png")

Bubble chart Network Analysis Discussion:

Running PANDA on TCGA and GTEx expression data, followed by GSEA, revealed the top targeted pathways in pancreatic cancer. The analysis identified aberrant signaling in pathways such as JAK/STAT signaling(Wang, 2021), TGF-β signaling, and the MAPK pathway, which have been previously linked to various cancer types. Additionally, changes in the activity of pathways like regulation of actin cytoskeleton, TCA cycle, and oxidative phosphorylation can serve as indicators of transformations commonly observed in different cancers. This network analysis highlights key pathways that may contribute to pancreatic cancer development and progression.

2.4 Single Gene Analysis

Upon conducting differential expression analysis comparing TCGA, that is, cancer samples, to the GTEx normal samples, using limma and voom, the plot of the final model mean-variance trend appeared not to display any concerning patterns, and the volcano plot indicated a large number of genes that are differentially expressed. A few of these differentially expressed genes will be discussed below.

Using the thresholds of a 0.05 adjusted p-value and a log fold change of 2, we found 3199 genes up-regulated in the Cancerous TCGA samples. Genes indicated as being upregulated included ANXA8 (logFC 6.08), FAM83A (logFC 8.52), KRT6A (logFC 3.27), MET (logFC 2.87), NT5E (logFC 4.48), and SLC2A1 (logFC 5.09). In a study(Posta et al., 2023) attempting to identify prognostic factors for pancreatic cancer, high expression of all of these genes were found to be linked with shorter survival. In addition, all of these genes were also found to affect this poor prognosis independently of each other. An overview of the function of these genes is provided in the table below, and the MET gene will be further discussed:

knitr::include_graphics("./single-gene-1.png")

The MET gene(NCBI, 2024) is located on the q arm of chromosome 7, and encodes a receptor for the Hepatocyte Growth Factor, upon binding to which the MET receptor activates pathways that regulate cell growth, survival, migration, and invasion. MET is a proto-oncogene that has been linked to multiple cancers, and specifically in pancreatic cancer, papers(Qin et al., 2022 & Lin et al., 2011 & Kim et al., 2024) suggest that the expression of MET is fundamental to the growth of the cancer cells, and increases the migration ability of cancer cells, leading to metastasis. MET has also been associated with cancer-related pain, via perineural invasion, and has been implicated in chemoresistance(Qin et al., 2022).

Using the thresholds of a 0.05 adjusted p-value and a log fold change of -2, we found 843 genes down-regulated in the Cancerous TCGA samples. Genes indicated as being downregulated included CPB1 (logFC -8.80), CPA1 (logFC -10.22), CPA2 (logFC -9.53), CTRB2 (logFC -9.60), and CTRC (-8.71). In a study(Xiao et al, 2022) attempting to identify genes relevant to drug development for the treatment of pancreatic cancer, these genes were found to be highly downregulated hub genes in protein-protein interaction networks, implicating these genes as critical in the tumor microenvironment. An overview of the function of these genes is provided in the table below, and the CTRB2 gene will be further discussed:

knitr::include_graphics("./single-gene-2.png")

The CTRB2 gene(NCBI, 2024) is located on the q arm of chromosome 16, and encodes a principal precursor involved in the activation of pancreatic enzymes, thus dysregulation of this gene results in imbalances in pancreatic enzyme activity, and inversions in this gene are associated with chronic inflammation of the pancreas. Specifically in relation to pancreatic cancer, low expression(Maturana et al., 2021) of CTRB2, via gene variations(Jermusyk et al., 2021) or otherwise, confers risk of cancer. The relationship between CTRB2 and type 2 diabetes(Maturana et al., 2021), as well as chronic pancreatitis, risk factors for pancreatic cancer, may further explain the association of this gene with pancreatic cancer.

3 Secondary Research Methods & Result: Male vs Female

3.1 Brief

Most of the steps are the same as in the TCGA vs GTEX analysis(analogous, similar purpose). Since this is our secondary research question, we have only drawn biological conclusions based on the overall analysis without examining detailed information such as individual KEGG, GO, and single gene analyses. Network Analysis, GO, KEGG, and DE were all performed, but with less interpretation compared to our primary research question (TCGA vs GTEX) due to limited time. Feel free to jump to the conclusion of this section if wanted.

3.2 EDA

3.2.1 Limma-Voom

# voom
condition_sex <- factor(colData(tcga_filtered)$tcga.gdc_cases.demographic.gender, 
                        levels=c("male", "female"))
design_sex <- model.matrix(~ condition_sex)
voomOutput_sex <- voom(tcga_exprs_filtered, design_sex, plot=TRUE)

knitr::include_graphics("./voom_sex.png")

3.2.2 Mean-Var trend

fit_sex <- lmFit(voomOutput_sex, design_sex)
fit_sex <- eBayes(fit_sex)
plotSA(fit_sex, main="Final model mean-variance trend")

knitr::include_graphics("./mean_var_sex.png")

ensembl <- useEnsembl(biomart = "genes", dataset = "hsapiens_gene_ensembl")
ensembl_ids <- row.names(res_sex[res_sex$adj.P.Val < 0.05 & abs(res_sex$logFC) > 1,])
ensembl_ids <- sub("\\..*", "", ensembl_ids)

# Query BioMart
gene_info <- getBM(
    attributes = c("ensembl_gene_id", "chromosome_name"),
    filters = "ensembl_gene_id",
    values = ensembl_ids,
    mart = ensembl
)

# Just prints which chromosome each gene is on
print(gene_info)

knitr::include_graphics("./de_chromosome.png")

3.2.3 PCA

tcga_exprs_filtered <- assay(tcga_filtered, "raw_counts")

# PCA - do we work here with results after filtering genes or not?
PCA <- prcomp(t(tcga_logtpm))
p <- autoplot(PCA, data = as.data.frame(colData(tcga)), 
              colour="tcga.gdc_cases.demographic.gender")
p

knitr::include_graphics("./pca_sex.png")

In the EDA section, the mean variance trend, limma-voom, both look fine as per lecture examples. The PCA is not as clearly separated between the two sets(Male vs Female) compared to the TCGA vs GTEx analysis.

3.3 GO Enrichment Analysis

3.3.1 Go: Over expressed in females

# Perform over-representation analyses for Gene Ontology terms using the limma package
go_females <- goana(up_females_ID, species="Hs")
# This is the list of top GO terms enriched for genes highly expressed in females
topGO(go_females, ontology = "BP")

knitr::include_graphics("./go_females.png")

3.3.2 Go: Over expressed in males

# Perform over-representation analyses for Gene Ontology terms using the limma package
go_males <- goana(up_males_ID, species="Hs")
# This is the list of top GO terms enriched for genes highly expressed in males
topGO(go_males, ontology = "BP")

knitr::include_graphics("./go_males.png")

3.4 Network Analysis

3.4.1 Gender Network

nodes       = data.frame(id = unique(as.vector(as.matrix(edges[,c(1,2)]))), 
                    label=unique(as.vector(as.matrix(edges[,c(1,2)]))))
nodes$group = ifelse(nodes$id %in% edges$from, "TF", "gene")
net <- visNetwork(nodes, edges, width = "100%")
net <- visGroups(net, groupname = "TF", shape = "triangle",
                 color = list(background = "purple", border="black"))
net <- visGroups(net, groupname = "gene", shape = "dot",       
                 color = list(background = "teal", border="black"))
visLegend(net, main="Legend", position="right", ncol=1) #GREEN is upregulated in cancer

knitr::include_graphics("./network_sex.png")

3.4.2 Kegg high in Male

library(tidyverse)
in_degree_male<- male_network_full |> group_by(gene) |>
  summarize(in_degree = sum(force))
in_degree_female<- female_network_full |> group_by(gene) |>
  summarize(in_degree = sum(force))

in_difference <- in_degree_male
in_difference$in_degree <- in_degree_male$in_degree - in_degree_female$in_degree

pathways <- gmtPathways("c2.cp.kegg_legacy.v2024.1.Hs.entrez.gmt")
in_difference$gene <- ifelse(annot[match(in_difference$gene, annot$ensembl_gene_id),"hgnc_symbol"]=="",in_difference$gene,  # If no match, keep Ensembl ID
  annot$entrezgene_id[match(in_difference$gene, annot$ensembl_gene_id)]  # Else, use HGNC symbol
)
in_difference <- in_difference[!is.na(in_difference$gene),]
in_difference <- in_difference[!duplicated(in_difference$gene), ]
in_difference <- setNames(in_difference$in_degree, in_difference$gene)

fgseaRes <- fgsea(pathways, in_difference, minSize=15, maxSize=500,nPermSimple = 10000)
sig <- fgseaRes[fgseaRes$padj < 0.05,]
sig[order(sig$NES)[1:10],] #Lower targeting in females, higher in males

knitr::include_graphics("./kegg_male.png")

3.4.3 Kegg high in Females

sig[order(-sig$NES)[1:10],] #Lower targeting in males, higher in females

knitr::include_graphics("./kegg_female.png")

3.4.4 PANDA by Gender

dat <- data.frame(fgseaRes)
# Settings
fdrcut <- 0.05 # FDR cut-off to use as output for significant signatures
dencol_neg <- "blue" # bubble plot color for negative ES
dencol_pos <- "red" # bubble plot color for positive ES
signnamelength <- 4 # set to remove prefix from signature names (2 for "GO", 4 for "KEGG", 8 for "REACTOME")
asp <- 3 # aspect ratio of bubble plot
charcut <- 100 # cut signature name in heatmap to this nr of characters
a <- as.character(dat$pathway) # 'a' is a great variable name to substitute row names with something more readable
for (j in 1:length(a)){
  a[j] <- substr(a[j], signnamelength+2, nchar(a[j]))
}
a <- tolower(a) # convert to lower case (you may want to comment this out, it really depends on what signatures you are looking at, c6 signatures contain gene names, and converting those to lower case may be confusing)
for (j in 1:length(a)){
  if(nchar(a[j])>charcut) { a[j] <- paste(substr(a[j], 1, charcut), "...", sep=" ")}
} # cut signature names that have more characters than charcut, and add "..."
a <- gsub("_", " ", a)
dat$NAME <- a
dat2 <- dat[dat[,"padj"]<fdrcut,]
dat2 <- dat2[order(dat2[,"padj"]),] 
dat2$signature <- factor(dat2$NAME, rev(as.character(dat2$NAME)))
sign_neg <- which(dat2[,"NES"]<0) 
sign_pos <- which(dat2[,"NES"]>0)
signcol <- rep(NA, length(dat2$signature))
signcol[sign_neg] <- dencol_neg # text color of negative signatures
signcol[sign_pos] <- dencol_pos # text color of positive signatures
signcol <- rev(signcol) # need to revert vector of colors, because ggplot starts plotting these from below
g<-ggplot(dat2, aes(x=padj,y=signature,size=size))
g+geom_point(aes(fill=NES), shape=21, colour="white")+
  theme_bw()+ # white background, needs to be placed before the "signcol" line
  xlim(0,fdrcut)+
  scale_size_area(max_size=10,guide="none")+
  scale_fill_gradient2(low=dencol_neg, high=dencol_pos)+
  theme(axis.text.y = element_text(colour=signcol))+
  theme(aspect.ratio=asp, axis.title.y=element_blank()) # test aspect.ratio

knitr::include_graphics("./panda_sex.png")

3.5 Gender Analysis Conclusion

Our analysis found 20 differentially expressed genes – 14 on the Y chromosome, 5 on the X chromosome, and 1 on a non-sex chromosome. The non-sex chromosome gene is a long intergenic non-protein coding RNA. The differences on the sex chromosomes were expected because of the gender comparison. The single gene on the non-sex chromosome is interesting and could be worth looking into further.

4 Secondary Research Methods & Result: Diabetic VS Non-Diabetic

4.1 Brief

Most of the steps are the same as in the TCGA vs GTEX analysis(analogous, similar purpose). Since this is our secondary research question, we have only drawn biological conclusions based on the overall analysis without examining detailed information such as individual KEGG, GO, and single gene analyses. Network Analysis, GO, KEGG, and DE were all performed, but with less interpretation compared to our primary research question (TCGA vs GTEX) due to limited time. Feel free to jump to the conclusion of this section if wanted.

4.2 EDA

4.2.1 Voom

condition_diabetes <- factor(tcga_colData_filtered_diab$tcga.xml_history_of_diabetes, 
                             levels=c("NO", "YES"))
design_diabetes <- model.matrix(~ condition_diabetes)
voomOutput_diabetes <- voom(tcga_exprs_filtered_diab, design_diabetes, plot=TRUE)

knitr::include_graphics("./voom_diabetes.png")

4.2.2 Mean-Variance

fit_diabetes <- lmFit(voomOutput_diabetes, design_diabetes)
fit_diabetes <- eBayes(fit_diabetes)
plotSA(fit_diabetes, main="Final model mean-variance trend")

knitr::include_graphics("./mean_diabetes.png")

4.2.3 PCA

knitr::include_graphics("./pca_diabetes.png")

In the EDA section, the mean variance trend, limma-voom, both look fine as per lecture examples. The PCA is not as clearly separated between the two sets(Diabetic vs Non-Diabetic) compared to the TCGA vs GTEx analysis.

4.3 Differentially Expressed Genes

knitr::include_graphics("./de_diabetes.png")

We found 0 differentially expressed genes in diabetic vs non-diabetic analysis.

4.4 Network Analysis

4.4.1 Network Graph

nodes     = data.frame(id = unique(as.vector(as.matrix(edges[,c(1,2)]))), 
                    label=unique(as.vector(as.matrix(edges[,c(1,2)]))))
nodes$group = ifelse(nodes$id %in% edges$from, "TF", "gene")
net <- visNetwork(nodes, edges, width = "100%")
net <- visGroups(net, groupname = "TF", shape = "triangle",
                 color = list(background = "purple", border="black"))
net <- visGroups(net, groupname = "gene", shape = "dot",       
                 color = list(background = "teal", border="black"))
visLegend(net, main="Legend", position="right", ncol=1)

knitr::include_graphics("./diabetes_graph.png")

4.4.2 KEGG low in diabetic

library(tidyverse)

in_degree_normal<- normal_network_full |> group_by(gene) |>
  summarize(in_degree = sum(force))
in_degree<- network_full |> group_by(gene) |>
  summarize(in_degree = sum(force))
in_degree <- in_degree[in_degree$gene %in% in_degree_normal$gene,]
in_difference <- in_degree_normal
in_difference$in_degree <- in_degree_normal$in_degree - in_degree$in_degree
pathways <- gmtPathways("c2.cp.kegg_legacy.v2024.1.Hs.entrez.gmt")
in_difference$gene <- ifelse(annot[match(in_difference$gene, annot$ensembl_gene_id),"hgnc_symbol"]=="",in_difference$gene,  # If no match, keep Ensembl ID
  annot$entrezgene_id[match(in_difference$gene, annot$ensembl_gene_id)]  # Else, use HGNC symbol
)
in_difference <- in_difference[!is.na(in_difference$gene),]
in_difference <- in_difference[!duplicated(in_difference$gene), ]
in_difference <- setNames(in_difference$in_degree, in_difference$gene)

fgseaRes <- fgsea(pathways, in_difference, minSize=15, maxSize=500,nPermSimple = 10000)
sig <- fgseaRes[fgseaRes$padj < 0.05,]
sig[order(sig$NES)[1:10],] #Lower targeting in diabetes

knitr::include_graphics("./kegg_diabetes_low.png")

4.4.3 Kegg high in diabetic

sig[order(-sig$NES)[1:10],] #Higher targeting in diabetes

knitr::include_graphics("./kegg_diabetes_high.png")

4.4.4 PANDA KEGG pathway Diabetes

dat <- data.frame(fgseaRes)
# Settings
fdrcut <- 0.05 # FDR cut-off to use as output for significant signatures
dencol_neg <- "blue" # bubble plot color for negative ES
dencol_pos <- "red" # bubble plot color for positive ES
signnamelength <- 4 # set to remove prefix from signature names (2 for "GO", 4 for "KEGG", 8 for "REACTOME")
asp <- 3 # aspect ratio of bubble plot
charcut <- 100 # cut signature name in heatmap to this nr of characters
a <- as.character(dat$pathway) # 'a' is a great variable name to substitute row names with something more readable
for (j in 1:length(a)){
  a[j] <- substr(a[j], signnamelength+2, nchar(a[j]))
}
a <- tolower(a) # convert to lower case (you may want to comment this out, it really depends on what signatures you are looking at, c6 signatures contain gene names, and converting those to lower case may be confusing)
for (j in 1:length(a)){
  if(nchar(a[j])>charcut) { a[j] <- paste(substr(a[j], 1, charcut), "...", sep=" ")}
} # cut signature names that have more characters than charcut, and add "..."
a <- gsub("_", " ", a)
dat$NAME <- a
dat2 <- dat[dat[,"padj"]<fdrcut,]
dat2 <- dat2[order(dat2[,"padj"]),] 
dat2$signature <- factor(dat2$NAME, rev(as.character(dat2$NAME)))
sign_neg <- which(dat2[,"NES"]<0) #Up regulated in cancer
sign_pos <- which(dat2[,"NES"]>0)
signcol <- rep(NA, length(dat2$signature))
signcol[sign_neg] <- dencol_neg # text color of negative signatures
signcol[sign_pos] <- dencol_pos # text color of positive signatures
signcol <- rev(signcol) # need to revert vector of colors, because ggplot starts plotting these from below
g<-ggplot(dat2, aes(x=padj,y=signature,size=size))
g+geom_point(aes(fill=NES), shape=21, colour="white")+
  theme_bw()+ # white background, needs to be placed before the "signcol" line
  xlim(0,fdrcut)+
  scale_size_area(max_size=10,guide="none")+
  scale_fill_gradient2(low=dencol_neg, high=dencol_pos)+
  theme(axis.text.y = element_text(colour=signcol))+
  theme(aspect.ratio=asp, axis.title.y=element_blank()) # test aspect.ratio

knitr::include_graphics("./panda_diabetes.png")

4.5 Diabetic Analysis Conclusion

Our analysis comparing diabetic and non-diabetic cancer groups found no differentially expressed genes with a log-fold change of 1 and an adjusted p-value of 0.05. Literature review however, indicated that type 2 diabetes has been associated with an almost twofold excess risk of pancreatic cancer, and that diabetes may be both caused by early pancreatic cancer and be a causal factor in the development of pancreatic cancer (Bosetti, 2014). Therefore, we proceeded with network analysis using PANDA, which showed some interesting differences in gene networks. We specifically chose to further investigate the following genes, PINX1, which was a positively targeted hub, and STC2, which was a negatively targeted hub.

The PINX1 gene is involved in a variety of complex cellular functions, and of relevance to cancer specifically, it acts to safeguard telomeres and chromosomal integrity, such that increased regulation of this gene in cancer may enhance cancer cell resistance to radiation and chemotherapy (You, 2024). Additionally, variations in this gene have been linked to risk of type two diabetes (Zee, 2011), indicating the differential targeting of this gene in those with pancreatic cancer with and without diabetes is of interest. Since the patterns of expression of PINX1 and the exact mechanisms of its influence in the development and progression of cancer differs by cancer type (You, 2024), further investigation is required to better understand the interplay between this gene, diabetes, and pancreatic cancer.

The STC2 gene is involved in cellular processes associated with glucose homeostasis and phosphorus metastasis (Lin, 2020), and has been found to be relevant to diabetes and cancer. For cancer specifically, this gene has been implicated in tumor development and progression, and has been noted as a prognostic predictor in cancers where overexpression predicted poor prognosis (Lin, 2020). However, the mechanism of the role of this gene in various cancers, especially in pancreatic cancer, is not well understood, with some studies indicating that expression of STC2 served as a tumor suppressor and inhibited cancer cell migration and invasion (Lin, 2020). Thus, similar to the PINX1 gene, further investigation is required to better understand the interplay between this gene, diabetes, and pancreatic cancer.

5 Secondary Research Methods & Result: Alcohol Consumption

5.1 Brief

Most of the steps are the same as in the TCGA vs GTEX analysis(analogous, similar purpose). Since this is our secondary research question, we have only drawn biological conclusions based on the overall analysis without examining detailed information such as individual KEGG, GO, and single gene analyses. Network Analysis, GO, KEGG, and DE were all performed, but with less interpretation compared to our primary research question (TCGA vs GTEX) due to limited time. Feel free to jump to the conclusion of this section if wanted.

5.2 EDA

5.2.1 Voom

condition_alcohol <- factor(tcga_colData_filtered_alcohol$tcga.gdc_cases.exposures.alcohol_history, 
                            levels=c("no", "yes"))
design_alcohol <- model.matrix(~ condition_alcohol)
voomOutput_alcohol <- voom(tcga_exprs_filtered_alcohol, design_alcohol, plot=TRUE)

knitr::include_graphics("./voom_alcohol.png")

5.2.2 Mean Var

fit_alcohol <- lmFit(voomOutput_alcohol, design_alcohol)
fit_alcohol <- eBayes(fit_alcohol)
plotSA(fit_alcohol, main="Final model mean-variance trend")

knitr::include_graphics("./mean_var_alcohol.png")

5.2.3 PCA alcohol

knitr::include_graphics("./pca_alcohol.png")

In the EDA section, the mean variance trend, limma-voom, both look fine as per lecture examples. The PCA is not as clearly separated between the two sets(Alcoholic vs Non-Alcohol) compared to the TCGA vs GTEx analysis.

5.3 Differentially Expressed Genes

knitr::include_graphics("./alcohol_DE.png")

We found 0 differentially expressed genes in the alcoholic vs non-alcoholic group.

5.4 Network Analysis

5.4.1 Network Graph

nodes     = data.frame(id = unique(as.vector(as.matrix(edges[,c(1,2)]))), 
                    label=unique(as.vector(as.matrix(edges[,c(1,2)]))))
nodes$group = ifelse(nodes$id %in% edges$from, "TF", "gene")
net <- visNetwork(nodes, edges, width = "100%")
net <- visGroups(net, groupname = "TF", shape = "triangle",
                 color = list(background = "purple", border="black"))
net <- visGroups(net, groupname = "gene", shape = "dot",       
                 color = list(background = "teal", border="black"))
visLegend(net, main="Legend", position="right", ncol=1) #GREEN is upregulated in alcohol people

knitr::include_graphics("./network_alcohol.png")

5.4.2 KEGG(Lower targeting in alcoholics)

library(tidyverse)
out_degree_normal<- normal_network_full |> group_by(tf) |>
  summarize(out_degree = sum(force))
out_degree<- network_full |> group_by(tf) |>
  summarize(out_degree = sum(force))
in_degree_normal<- normal_network_full |> group_by(gene) |>
  summarize(in_degree = sum(force))
in_degree<- network_full |> group_by(gene) |>
  summarize(in_degree = sum(force))
in_degree <- in_degree[in_degree$gene %in% in_degree_normal$gene,]
in_difference <- in_degree_normal
in_difference$in_degree <- in_degree_normal$in_degree - in_degree$in_degree

pathways <- gmtPathways("c2.cp.kegg_legacy.v2024.1.Hs.entrez.gmt")
in_difference$gene <- ifelse(annot[match(in_difference$gene, annot$ensembl_gene_id),"hgnc_symbol"]=="",in_difference$gene,  # If no match, keep Ensembl ID
  annot$entrezgene_id[match(in_difference$gene, annot$ensembl_gene_id)]  # Else, use HGNC symbol
)
in_difference <- in_difference[!is.na(in_difference$gene),]
in_difference <- in_difference[!duplicated(in_difference$gene), ]
in_difference <- setNames(in_difference$in_degree, in_difference$gene)

fgseaRes <- fgsea(pathways, in_difference, minSize=15, maxSize=500,nPermSimple = 10000)
sig <- fgseaRes[fgseaRes$padj < 0.05,]
sig[order(sig$NES)[1:10],] #Lower targeting in alcoholics

knitr::include_graphics("./lower_alcohol.png")

5.4.3 KEGG(Higher targeting in alcoholics)

sig[order(-sig$NES)[1:10],] #higher  targeting in alcoholics

knitr::include_graphics("./higher_alcohol.png")

5.4.4 PANDA

dat <- data.frame(fgseaRes)
# Settings
fdrcut <- 0.05 # FDR cut-off to use as output for significant signatures
dencol_neg <- "blue" # bubble plot color for negative ES
dencol_pos <- "red" # bubble plot color for positive ES
signnamelength <- 4 # set to remove prefix from signature names (2 for "GO", 4 for "KEGG", 8 for "REACTOME")
asp <- 3 # aspect ratio of bubble plot
charcut <- 100 # cut signature name in heatmap to this nr of characters
a <- as.character(dat$pathway) # 'a' is a great variable name to substitute row names with something more readable
for (j in 1:length(a)){
  a[j] <- substr(a[j], signnamelength+2, nchar(a[j]))
}
a <- tolower(a) # convert to lower case (you may want to comment this out, it really depends on what signatures you are looking at, c6 signatures contain gene names, and converting those to lower case may be confusing)
for (j in 1:length(a)){
  if(nchar(a[j])>charcut) { a[j] <- paste(substr(a[j], 1, charcut), "...", sep=" ")}
} # cut signature names that have more characters than charcut, and add "..."
a <- gsub("_", " ", a)
dat$NAME <- a
dat2 <- dat[dat[,"padj"]<fdrcut,]
dat2 <- dat2[order(dat2[,"padj"]),] 
dat2$signature <- factor(dat2$NAME, rev(as.character(dat2$NAME)))
sign_neg <- which(dat2[,"NES"]<0) #Up regulated in cancer
sign_pos <- which(dat2[,"NES"]>0)
signcol <- rep(NA, length(dat2$signature))
signcol[sign_neg] <- dencol_neg # text color of negative signatures
signcol[sign_pos] <- dencol_pos # text color of positive signatures
signcol <- rev(signcol) # need to revert vector of colors, because ggplot starts plotting these from below
g<-ggplot(dat2, aes(x=padj,y=signature,size=size))
g+geom_point(aes(fill=NES), shape=21, colour="white")+
  theme_bw()+ # white background, needs to be placed before the "signcol" line
  xlim(0,fdrcut)+
  scale_size_area(max_size=10,guide="none")+
  scale_fill_gradient2(low=dencol_neg, high=dencol_pos)+
  theme(axis.text.y = element_text(colour=signcol))+
  theme(aspect.ratio=asp, axis.title.y=element_blank()) # test aspect.ratio

knitr::include_graphics("./panda_maotai.png")

5.5 Alcoholic Analysis Conclusion

Our analysis found no differentially expressed genes related to alcohol. While network analysis was conducted, we did not investigate it further. This area could be interesting for future research to uncover potential hidden mechanisms.

6 Overall Limitations & Future Directions

For our TCGA vs GTEx study:

Some of the limitations of our study include a relatively small sample size, with 349 samples from GTEx and 179 samples from TCGA. This potentially reduces the reliability of our study because pancreatic cancer, although one of the deadliest, is rare. It would be beneficial to include more GTEx and TCGA samples to improve our understanding, as genetic variation can be significant among individuals in TCGA and GTEx as well. Our study primarily focuses on the early stages (1-2) of pancreatic cancer, thereby excluding information from the later stages (3-5), which could potentially provide more comprehensive insights into the overall cancer mechanism when compared to GTEx.

For the future, we plan to address the limitations mentioned above. We aim to include genetic information from later stages of cancer for a more thorough analysis, allowing us to compare stages 1-5 with stages 1-2 for a broader perspective. Additionally, we can try to sample more TCGA data, similar to GTEx, to reduce the effects of genetic variations between individuals.

For our TCGA diabetes vs non-diabetes analysis, we identified no differentially expressed genes, but the underlying mechanisms are not well understood. Some effects may be driven by diabetes itself rather than cancer, making it difficult to separate the two. This highlights the need for further investigation to clarify these relationships and improve the accuracy of our findings.

For the gender analysis, a limitation of this would be the strong bias towards sex chromosomes, which may overlook important gene expression differences on autosomal chromosomes. For future direction, increasing the sample size and refining the analysis to detect more subtle autosomal differences could provide a clearer understanding of gene expression variations between males and females. Also, perform analysis when accounting for only non-sex chromosome might lead to more meaningful results, where we can elminate the noise from X,Y chromosomes.

For the alcoholic analysis, a limitation is the varying effects of alcohol, where some low-dosage drinker were also accounted as in the alcoholic group. But it might have very minimal effect towards pancreatic function. We believe to account only for high-dosage alcohol consumer vs non-alcoholic group, accompanying with a subsequent network analysis for this variable may reveal more interesting patterns.

7 References

Bosetti, C., Rosato, V., Li, D., Silverman, D., Petersen, G. M., Bracci, P. M., Neale, R. E., Muscat, J., Anderson, K., Gallinger, S., Olson, S. H., Miller, A. B., Bas Bueno-de-Mesquita, H., Scelo, G., Janout, V., Holcatova, I., Lagiou, P., Serraino, D., Lucenteforte, E., & Fabianova, E. (2014). Diabetes, antidiabetic medications, and pancreatic cancer risk: an analysis from the International Pancreatic Cancer Case-Control Consortium. Annals of Oncology: Official Journal of the European Society for Medical Oncology, 25(10), 2065–2072. https://doi.org/10.1093/annonc/mdu276‌

Chen, D., Zhang, S., & Fang, W. (2023). A single-cell atlas of the tumor, stromal and immune microenvironments in pancreatic cancer liver metastasis (GSE197177) [Data set]. NCBI Gene Expression Omnibus. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE197177

Dafrazi, A. A., Mehrabi, T., & Malekinejad, F. (2023). A bioinformatics study for recognition of hub genes and pathways in pancreatic ductal adenocarcinoma. arXiv preprint arXiv:2303.14440. https://doi.org/10.48550/arXiv.2303.14440

Jermusyk, A., Zhong, J., Connelly, K. E., Petersen, G. M., Westlake, C. J., & Amundadottir, L. T. (2021). A 584 bp deletion in CTRB2 inhibits chymotrypsin B2 activity and secretion and confers risk of pancreatic cancer. The American Journal of Human Genetics, 108(10), 1885–1898. https://doi.org/10.1016/j.ajhg.2021.09.012

Kim, J., Lee, T. S., Lee, M. H., Cho, I. R., Ryu, J. K., Kim, Y. T., Lee, S. H., & Paik, W. H. (2024). Pancreatic cancer treatment targeting the HGF/c-MET pathway: The MEK inhibitor trametinib. Cancers, 16(5), 1056. https://doi.org/10.3390/cancers16051056

Li, C., Wu, J.-J., Hynes, M., Welling, T. H., Pasca di Magliano, M., & Simeone, D. M. (2011). c-Met is a marker of pancreatic cancer stem cells and therapeutic target. Gastroenterology, 141(6), 2218–2227. https://doi.org/10.1053/j.gastro.2011.07.050

Lin, C., Sun, L., Huang, S., Weng, X., & Wu, Z. (2020). STC2 is a potential prognostic biomarker for pancreatic cancer and promotes migration and invasion by inducing epithelial–mesenchymal transition. Frontiers in Oncology, 10, 290. https://doi.org/10.3389/fonc.2020.00290

López de Maturana, E., Rodríguez, J. A., Alonso, L., Lao, O., Molina-Montes, E., Martín-Antoniano, I. A., Gómez-Rubio, P., Lawlor, R., Carrato, A., Hidalgo, M., Iglesias, M., Molero, X., Löhr, M., Michalski, C., Perea, J., O’Rorke, M., Barberà, V. M., Tardón, A., Farré, A., … Malats, N. (2021). A multilayered post-GWAS assessment on genetic susceptibility to pancreatic cancer. Genome Medicine, 13(15). https://doi.org/10.1186/s13073-020-00816-4

National Center for Biotechnology Information (NCBI). (2024). CTRB2 chymotrypsinogen B2 [Homo sapiens (human)]. National Library of Medicine (NLM). Retrieved from https://www.ncbi.nlm.nih.gov/gene/440387

National Center for Biotechnology Information (NCBI). (2024). MET proto-oncogene, receptor tyrosine kinase [Homo sapiens (human)]. National Library of Medicine (NLM). Retrieved from https://www.ncbi.nlm.nih.gov/gene/4233

Posta, M., & Győrffy, B. (2023). Analysis of a large cohort of pancreatic cancer transcriptomic profiles to reveal the strongest prognostic factors. Clinical and Translational Science, 16(8), 1479–1491. https://doi.org/10.1111/cts.13563

Qin, T., Xiao, Y., Qian, W., Wang, X., Gong, M., Wang, Q., An, R., Han, L., Duan, W., Ma, Q., & Wang, Z. (2022). HGF/c-Met pathway facilitates the perineural invasion of pancreatic cancer by activating the mTOR/NGF axis. Cell Death & Disease, 13(4). https://doi.org/10.1038/s41419-022-04799-5

UC Davis Bioinformatics Core. (2018, June). Differential expression with limma-voom. Retrieved from https://ucdavis-bioinformatics-training.github.io/2018-June-RNA-Seq-Workshop/thursday/DE.html

Wang, S., Zheng, Y., Yang, F., Zhu, L., Zhu, X.-Q., Wang, Z.-F., Wu, X.-L., Zhou, C.-H., Yan, J.-Y., Hu, B.-Y., Kong, B., Fu, D.-L., Bruns, C., Zhao, Y., Qin, L.-X., & Dong, Q.-Z. (2021). The molecular biology of pancreatic adenocarcinoma: Translational challenges and clinical perspectives. Signal Transduction and Targeted Therapy, 6(1), 249. https://doi.org/10.1038/s41392-021-00659-4

Wikipedia contributors. (n.d.). Immune system. Wikipedia, The Free Encyclopedia. Retrieved December 25, 2024, from https://en.wikipedia.org/wiki/Immune_system

Wikipedia contributors. (n.d.). Tumor metabolome. Wikipedia, The Free Encyclopedia. Retrieved October 29, 2024, from https://en.wikipedia.org/wiki/Tumor_metabolome

Xiao, X., Wan, Z., Liu, X., Chen, H., Zhao, X., Ding, R., Cao, Y., Zhou, F., Qiu, E., Liang, W., Ou, J., Chen, Y., Chen, X., & Zhang, H. (2023). Screening of therapeutic targets for pancreatic cancer by bioinformatics methods. International Journal of Gastrointestinal Cancer, 55(6), 420–425. https://doi.org/10.1055/a-2007-2715

Xiao, Y., Zhang, B., Cloyd, J. M., Xu, G., Du, S., Mao, Y., & Pawlik, T. M. (2022). Gene signature and connectivity mapping to assist with drug prediction for pancreatic ductal adenocarcinoma. Surgical Oncology, 44, 101849. https://doi.org/10.1016/j.suronc.2022.101849

You, D., Tong, K., Li, Y., Zhang, T., Wu, Y., Wang, L., Chen, G., & Zhang, X. (2024). PinX1 plays multifaceted roles in human cancers: A review and perspectives. Molecular Biology Reports, 51(1), 1163. https://doi.org/10.1007/s11033-024-10082-x

Zee Robert Y. L., Ridker, P. M., & Chasman, D. I. (2011). Genetic variants of 11 telomere-pathway gene loci and the risk of incident type 2 diabetes mellitus: The Women’s Genome Health Study. Atherosclerosis, 218(1), 144–146. https://doi.org/10.1016/j.atherosclerosis.2011.05.013